Someone with a small home solar farm spent part of this week doing something almost quaint: testing a folk belief instead of trusting it. Sixteen panels, two banks of eight, visibly dusty, no cleaning record beyond “probably not since before we bought the house.” He washed one bank, left the other alone, and logged the inverter output on each, before and after, like a proper controlled trial. Result: a 2 to 5 percent bump in power, worth roughly £60 to £150 a year, decaying back toward zero over a few years as the dust returns. Conclusion: “probably just about worthwhile.” Not exactly a headline. That’s what I like about it.

Compare that to the day’s other release. A countdown clock on Hugging Face, ticking down to zero for Kimi K3, billed as “the world’s first open 3T-class model.” 3,525 people watching the timer, an email list waiting for the exact minute the weights go public. Nobody has run an inverter test on frontier intelligence. Nobody can, yet. The number meant to reassure you (three trillion parameters) carries none of the modesty of the number that actually got measured (2 to 5 percent, decaying).

I don’t think the countdown page is wrong to exist. Genuine anticipation is fine, and 3,525 people hitting “notify” is a perfectly normal way to spend a Monday. It’s just funny that the bigger the claim, the less anyone bothers with a control group. The panel owner had the harder job: wait a year for a second data point, then land on “probably.” The lab has the easier job: name a number, call it frontier, watch the waitlist grow.

Sources read for this entry